slewis

The parent comment does address the points, by saying that while they are downsides, the new properties of crypto _might_ outweigh them. It is nuanced, unlike the comment I am replying to.

Refactoring is much easier in a well typed codebase.

When refactoring is easier, it happens more often, which means code quality and readability are higher, which reduces the likelihood of every other kind of error.

I stumbled on this excellent article from 2015 that describes a really cool mechanism for tracing function execution in Python. This technique is used in the [autograd library](https://github.com/HIPS/autograd), which is a precursor to [Jax](https://github.com/google/jax), a competitor to Tensorflow and pytorch that's starting to gain traction.

They’re arguing: better for developers is better for users. Developers get to spend more time on building features and less on the accidental complexity that comes with supporting all of the competing platforms and frameworks.

For most purposes you don’t need to train from scratch. Instead you fine-tune an existing model, on smaller amounts of data and for a fraction of the time.

This is akin to teaching an adult human about a specific domain. Better to just do that than make a whole new human from scratch!

You’re expressing your opinion as if it’s a cold hard truth.

Love this quote from Roger Ebert. Thanks for posting it. He was a great reviewer because he loved what he did. His passion came through in everything he wrote.

It’s Saturday now in many places in the world. Can you move a few thousand USD between your two primary accounts at different traditional banks before Monday?

The article is paywalled. So people upvoting this, are you just upvoting for the abstract and headline alone?

What did you ask?

I recommend just treating undefined and null as equivalent (since you can't control which one or ones third-party libraries will use). Everywhere you need to check for them do `!= null` or `== null` which is an idiomatic way to check for either.

"you can already have fast and scalable transactions through normal banking."

I see this repeated a lot. I think you're saying bitcoin is no better than normal banking for moving value.

Do you really believe this? Yes, in the US I can immediately wire money somewhere as long as:

1) I do it within east coast business hours. Weekend? Sorry, wait til Monday 2) I have the correct information for the receiving institution, some of which require a small transfer up front to confirm the setup. 3) My bank actually supports wiring (Wealthfront, for example, only supports ACH transfers which take a few days) 4) I'm sending within the US, otherwise there are a bunch of other complications and requirements

I also need to trust that the bank actually has my money.

With cryptocurrency you can confirm that your stored value exists and is stilled owned by you, and transfer it to anywhere else within 30 minutes, in a provably secure and verifiable way.

As others have mentioned, you can’t get rid of null, because of third-party libraries for one. Instead your code should treat them the same at comparison time: ‘== null’ is the canonical way to check for either.

FWIW, our Artifacts tool can track references to data that exists in other systems, or store the data directly which allows us to ensure there are no "dark matter" users like you mentioned. It's a flexible system.

Shameless plug: my company Weights & Biases (https://wandb.com) has tackled this in a different way. We make tools to keep track of results across your pipelines regardless of how you choose to orchestrate execution. This is one of our big selling points: very simple on-ramp to reproducible tracking, and infrastructure agnosticism. These have led to broad adoption across companies and academics. We also have a pretty cool UI.

Pachyderm makes different tradeoffs and we're excited to see their launch. Seriously congrats to you all! This is an invigorating space to say the least ;).

I think novelty of a moment relative to your recorded memories affects how you perceive the duration of that moment. More novel moments are perceived as lasting longer.

As life goes on novelty of new moments goes down.

You could say this explains or underlies the Janet model.

But one can seek out novelty to slow down time. In the extreme you could “reset”, move away from everything you know, and start from scratch.

I have no evidence for any of this other than my own perception. Does anyone know where to find more discussion on this topic?

It’s interesting that folks are generally giving Trump the benefit of the doubt. There are many comments to the effect of “while there is no public evidence, maybe the US government has secret evidence that TikTok has broken the law.”

Trump is incredibly vindictive. Remember a month ago when he hosted a way under-attended rally, because of a fake registration movement that spread on, yup, TikTok? Trump remembers.

Sounds like you’re not convinced.

It's fair to call out Atlassian & Jira here. But I want to give them some credit. They recently rolled out a Jira change that was a huge step in the right direction (navbar at the top that can take you directly to projects and commonly used filters, among other things). It is a HUGE improvement. So it seems they may have finally figured out how to listen and improve UX. This is no small task for software that's used by so many different orgs for as many different purposes.

But really what holds it back still is sluggishness. I'd take a few extra clicks in exchange for instantaneous responses to each click.

In reinforcement learning, you typically need to run many copies of an environment (an Atari emulator for example) to collect experience for your agents. Environment code usually makes more sense on a CPU, because it’s a lot more branchy and stateful.

Think of two agents that run for awhile and end up in different parts of a virtual world, that require very different code to execute. This might be pretty difficult to parallelize on a GPU.

After collecting a bunch of experience, you update whatever function you’re optimizing. This could be something like a neural network that is the brain of your agents. The update step can happen on a GPU.

But then we need to go collect a lot more experience again, so we’re back in CPU land.

Experience collection often dominates overall compute time in reinforcement learning. Which means you want a lot of CPUs.

grep

Did your business get off the ground?

Yes. It’s built by Microsoft and has huge adoption. Typescript is here to stay.

If we can assume determinism, there’s a much better compression algorithm.

Cool! Can you provide some real world examples of issues you flag? I poked around on your site and didn’t see any.

We always look at number of stars or installs of libraries before deciding to put them in a production project. It'd be cool if there was a "percent unsafe" badge that was commonly used for Rust repos. If folks started paying attention to it it'd be harder for unsafe libraries to become popular.

There's been a lot of thought at Weights & Biases about the tradeoff between competition and collaboration. Competition certainly fosters activity, but it encourages behaviors like ensembling everything to eek out a few more hundredths of a percent. This benchmark isn't incentivized in any way other than "let's drive the field forward", so we may see less of that behavior.

We've considered benchmarks that proceed in phases: a closed competitive phase for 3 months, then award a prize to the top result, and another prize for best user writeup. Follow that by a collaborative phase where it's more about sharing, teamwork etc. Rinse and repeat.

The question of attribution is really interesting. Who made the largest contribution to the development of a model? It could have actually happened in a hallway conversation, or something equally as untrackable. We'd love to hear other peoples' thoughts on this.

Stacey on our team has put a lot of thought into these topics and may have more to say here!

Shawn from Weights & Biases here. We've been working with Github and Microsoft Research on this for just about a year now and we're super excited to launch it today.

We've seen huge advances in human language modeling and translation due to the success of deep learning. Often new directions start with a really motivated team producing a new kind of dataset. Who better to do that for code as language than Github!

This started as a grassroots effort inside of Github, and went through many iterations. When it was presented to Github's CEO six months ago, he correctly pointed out that we needed to go back and include Github's most popular language (javascript). As the project went on many smart people chipped in, and we produced something that we think is truly useful.

Check out the paper here: https://arxiv.org/abs/1909.09436

We overcame plenty of challenges to pull this off. For example: how do you clean this data? how do you label it? We've got folks from Github, Microsoft Research and Weights & Biases here to answer any and all questions you might have. Can't wait to see where this goes!

Economies of scale might give the market owner such a huge advantage in food delivery logistics (coupled with ride logistics etc), that it’s worth losing a lot until inflection. The winner forces everyone out, and gets there by spending more up front.